CN109754317A - Merge interpretation clothes recommended method, system, equipment and the medium of comment - Google Patents
Merge interpretation clothes recommended method, system, equipment and the medium of comment Download PDFInfo
- Publication number
- CN109754317A CN109754317A CN201910024347.2A CN201910024347A CN109754317A CN 109754317 A CN109754317 A CN 109754317A CN 201910024347 A CN201910024347 A CN 201910024347A CN 109754317 A CN109754317 A CN 109754317A
- Authority
- CN
- China
- Prior art keywords
- picture
- jacket
- lower clothing
- visual signature
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present disclosure discloses interpretation clothes recommended method, system, equipment and the media of fusion comment, comprising: coder-decoder neural network model of the building based on deep learning；Coder-decoder neural network model based on deep learning is trained；By jacket picture and lower clothing picture to be recommended, it is input to trained coder-decoder neural network model simultaneously, the model gives a mark to the matching degree of jacket picture and lower clothing picture, provides recommendation results according to marking sequence, while providing the simulation comment of matching degree.It trains recommended models using the useful information hidden in user comment, to promote the effect of recommendation, while energy analog subscriber generates explanation of the comment as recommendation to the result recommended, and improves the interpretation of recommendation.
Description
Technical field
This disclosure relates to which clothes recommend field, more particularly to merges the interpretation clothes recommended method of comment, system, sets
Standby and medium.
Background technique
The statement of this part is only to refer to background technique relevant to the disclosure, not necessarily constitutes the prior art.
The purpose that clothes are recommended is to promote people to online purchase by recommending possible interested fashion clothing to user
The interest and participation of object.On the one hand clothes recommended technology can help user quickly to search in dazzling online swank
The clothes of rope oneself satisfaction, on the other hand can also help online retailer to improve service quality and expand income.Therefore, it takes now
Dress recommended technology plays more and more important role in online retail market, also results in the extensive of industry and academia
Pay attention to.
Clothes recommend field to contain many problems, the present invention towards particular problem be the jacket given for user
(such as T-shirt, housing) is recommended suitably to descend clothing (such as skirt, shorts), and vice versa.The solution of the problem can help to use
Family is preferably arranged in pairs or groups oneself clothes, and user is allowed to become more fashion.The number that the clothes of early stage recommend research to mark based on expert
According to collection, these data sets are mostly too small, limit the exploitation (for example, model based on deep learning) of complex model.In recent years, with
The appearance of the on-line communities (such as Polyvore and Chictopia) of fashion guiding, people can share and comment on clothes and take
Match.In addition to a large amount of garment coordination, these also include that other valuable information (such as a large amount of are used from data of crowdsourcing
Family comment), it can be used for constructing more acurrate and intelligent recommender system.
Current clothes recommended technology is mostly merely sentenced dependent on visual signature is extracted from the picture of jacket and lower clothing
Matching degree between the lower clothing of the jacket and candidate that break given.They all ignore the information in user comment, without reference to
The comment of user is to learn the matching grating between general clothes.Current simultaneously clothes recommended technology more provide a judgement
As a result, generating the reasons why comment is recommended without analog subscriber.This to recommend to lack clarity and accountability.
Summary of the invention
In order to solve the deficiencies in the prior art, present disclose provides the interpretation clothes recommended method of fusion comment, it is
System, equipment and medium, train recommended models using the useful information hidden in user comment, to promote the effect of recommendation
Fruit, while energy analog subscriber generates explanation of the comment as recommendation to the result recommended, and improves the interpretation of recommendation.
In a first aspect, present disclose provides the interpretation clothes recommended methods of fusion comment；
Merge the interpretation clothes recommended method of comment, comprising:
Construct the coder-decoder neural network model based on deep learning；
Coder-decoder neural network model based on deep learning is trained；
By jacket picture and lower clothing picture to be recommended, while being input to trained coder-decoder neural network
Model, the model give a mark to the matching degree of jacket picture and lower clothing picture, provide recommendation results according to marking sequence, together
When provide matching degree simulation comment.
As a kind of possible implementation, the coder-decoder neural network model based on deep learning, packet
It includes:
Jacket encoder, lower clothing encoder, matching and decoding device and generation decoder；
The jacket encoder extracts the jacket visual signature and jacket coding of jacket picture for receiving jacket picture
It indicates；The jacket coded representation includes the match information between jacket picture and lower clothing picture；
The lower clothing encoder extracts the lower clothing visual signature and lower clothing coding of lower clothing picture for receiving lower clothing picture
It indicates；The lower clothing coded representation includes the match information between jacket picture and lower clothing picture；
The matching and decoding device is used for according to jacket coded representation and lower clothing coded representation, to jacket picture and lower clothing picture
Between matching degree give a mark；
The generation decoder is used to be compiled according to jacket visual signature, jacket coded representation, lower clothing visual signature and lower clothing
Code indicates, simulates comment to the combination producing of jacket picture and lower clothing picture.
As a kind of possible implementation, the specific steps of the jacket visual signature of jacket picture are extracted are as follows:
The jacket encoder, comprising: sequentially connected first convolutional layer, the second convolutional layer, the first splicing layer and the
One pond layer；
First convolutional layer carries out Visual Feature Retrieval Process to jacket picture, obtains First look feature；
Second convolutional layer carries out Visual Feature Retrieval Process to jacket picture, obtains the second visual signature；
The first splicing layer carries out series connection splicing to First look feature and the second visual signature, that splicing is obtained
Three visual signatures are sent into the first pond layer；
First pond layer handles third visual signature, obtains the jacket visual signature of jacket picture.
As a kind of possible implementation, the specific steps of the lower clothing visual signature of lower clothing picture are extracted are as follows:
The lower clothing encoder, comprising: sequentially connected third convolutional layer, Volume Four lamination, the second splicing layer and the
Two pond layers；
The third convolutional layer carries out Visual Feature Retrieval Process to lower clothing picture, obtains the 4th visual signature；
The Volume Four lamination carries out Visual Feature Retrieval Process to lower clothing picture, obtains the 5th visual signature；
The second splicing layer carries out series connection splicing to the 4th visual signature and the 5th visual signature, by what is obtained after splicing
6th visual signature is sent into the second pond layer；
Second pond layer handles the 6th visual signature, obtains the visual signature of lower clothing picture.
As a kind of possible implementation, the specific steps of the jacket coded representation of jacket picture are extracted are as follows:
Using interaction attention mechanism, the match information between jacket picture and lower clothing picture is encoded to the jacket of extraction
In the visual signature of picture, the coded representation of jacket picture is obtained.
As a kind of possible implementation, the specific steps of the lower clothing coded representation of lower clothing picture are extracted are as follows:
Using interaction attention mechanism, the match information between jacket picture and lower clothing picture is encoded to the lower clothing of extraction
In the visual signature of picture, the coded representation of lower clothing picture is obtained.
It will be between jacket picture and lower clothing picture using interaction attention mechanism as a kind of possible implementation
Match information is encoded in the visual signature of jacket picture of extraction, obtains the specific steps of the coded representation of jacket picture are as follows:
Firstly, obtaining the global characteristics of lower clothing picture by the average value for calculating lower clothing visual signature；
Then, to each visual signature of jacket picture, vision of the global characteristics to jacket picture of lower clothing picture is calculated
The attention weight of feature；Attention weight is normalized；
Secondly, using the global characteristics of lower clothing picture to the attention weight of the visual signature of jacket picture, to jacket figure
The visual signature of piece is weighted summation, obtains the attention global characteristics of jacket picture；
Again, the attention global characteristics of jacket picture are mapped to visual feature vector；
Again, the visual feature vector of jacket picture jacket article vector corresponding with jacket picture is subjected to series connection spelling
It connects, the result being spliced to is the coded representation of final jacket picture.
As a kind of possible implementation, the obtaining step of jacket article vector are as follows:
Firstly, one jacket article vector matrix of random initializtion, the corresponding jacket of each of these row；
Then, according to the jacket picture of input, corresponding vector is obtained from jacket article vector matrix, for later
It calculates；
Finally, jacket article vector matrix, with the minimum target of loss function value, is led to together with the parameter of neural network
It crosses back-propagating BP algorithm to be updated, finally obtains updated jacket article vector.
Jacket article vector captures the useful information in history match record by back-propagating BP algorithm as to upper
The supplement of clothing visual signature.
The random initializtion using Xavier method, be uniformly distributed or one of normal distribution method.
It will be between jacket picture and lower clothing picture using interaction attention mechanism as a kind of possible implementation
Match information is encoded in the visual signature of lower clothing picture of extraction, obtains the specific steps of the coded representation of lower clothing picture are as follows:
Firstly, obtaining the global characteristics of jacket picture by the average value for calculating jacket visual signature；
Then, to each visual signature of lower clothing picture, vision of the global characteristics to lower clothing picture of jacket picture is calculated
It is special
The attention weight of sign；Attention weight is normalized；
Secondly, using the global characteristics of jacket picture to the attention weight of the visual signature of lower clothing picture, to lower clothing figure
The visual signature of piece is weighted summation, obtains the attention global characteristics of lower clothing picture；
Again, the attention global characteristics of lower clothing picture are mapped to visual feature vector；
Again, the visual feature vector of lower clothing picture lower clothing product vector corresponding with lower clothing picture is subjected to series connection spelling
It connects, the result being spliced to is the coded representation of final lower clothing picture.
As a kind of possible implementation, the obtaining step of lower clothing product vector are as follows:
Firstly, random initializtion one lower clothing product vector matrix, the corresponding lower clothing of each of these row；
Then, according to the lower clothing picture of input, corresponding vector is obtained from lower clothing product vector matrix, for later
It calculates；
Finally, lower clothing product vector matrix, with the minimum target of loss function value, is led to together with the parameter of neural network
It crosses back-propagating BP algorithm to be updated, finally obtains updated lower clothing product vector.
Lower clothing product vector captures the useful information in history match record by back-propagating BP algorithm as under
The supplement of clothing visual signature.
As a kind of possible implementation, according to jacket coded representation and lower clothing coded representation, to jacket picture under
The specific steps that matching degree between clothing picture is given a mark are as follows:
Using jacket coded representation and lower clothing coded representation as input value, it is input in MLP multi-layer perception (MLP), output is
The matching marking result of jacket picture and lower clothing picture.
As a kind of possible implementation, according to jacket visual signature, jacket coded representation, lower clothing visual signature and under
Clothing coded representation simulates the step of commenting on to the combination producing of jacket picture and lower clothing picture are as follows:
Step (1): building gating cycle neural network GRU；
Step (2): using the coded representation of jacket and lower clothing, the original state of gating cycle neural network GRU is calculated；
Step (3): gating cycle neural network GRU carries out the circulate operation of step (31) to step (33) until generating one
A complete sentence:
Step (31): it first to jacket visual signature and lower clothing visual signature, is handled using cross-module state attention mechanism
Obtain the context vector of current time step；
Step (32): by the state of a upper time step of gating cycle neural network GRU, a upper time step generates word
Term vector and current time step context vector be input in gating cycle neural network GRU, obtain current time step
New state and to be currently generated word prediction probability be distributed；
Step (33): the word of maximum probability is chosen as current generation result；The word includes punctuation mark；Such as
The current generation result of fruit is fullstop, illustrates to have generated a complete sentence, the then word generated all time steps
A sentence is sequentially connected into return.
Cross-module state attention is utilized to jacket visual signature and lower clothing visual signature as a kind of possible implementation
Mechanism is handled to obtain the specific steps of context vector are as follows:
Firstly, corresponding tandem compound to jacket visual signature and lower clothing visual signature；
Then, the state of the upper time step of gating cycle neural network GRU is calculated to jacket visual signature and lower clothing
After visual signature tandem compound, each combined attention weight；
Then, using the attention weight of calculating, weighted sum is done to all tandem compounds, the result finally returned to is just
It is the context vector of current time step.
As a kind of possible implementation, the acquisition modes of term vector are:
Firstly, one term vector matrix of random initializtion, the corresponding word of each of these row；
Then, according to word currently entered, corresponding vector is obtained from term vector matrix, for calculating later；
Finally, term vector matrix is by together with the parameter of neural network, with the minimum target of loss function, by rear to biography
BP algorithm is broadcast to be updated.
As a kind of possible implementation, the coder-decoder neural network model based on deep learning is carried out
Trained specific steps are as follows:
Training set includes the matched jacket that the real user crawled from online fashion community website provides and lower clothing
Combination, each combination include jacket picture, lower clothing picture, thumb up several and user comment；
Combination of the number greater than threshold value will be thumbed up and be considered as matching combination；Then combination is mismatched to obtain by negative sampling, i.e.,
It randomly selects a jacket and a lower clothing constitutes a combination, if the combination does not occur in matching combination, by the group
Conjunction is considered as mismatch combination；To matching combination in jacket picture and lower clothing picture, respectively extract jacket picture visual signature,
The coded representation of jacket picture, the coded representation of the visual signature of lower clothing picture and lower clothing picture；
To the jacket picture and lower clothing picture mismatched in combination, visual signature, the jacket figure of jacket picture are extracted respectively
The coded representation of piece, the coded representation of the visual signature of lower clothing picture and lower clothing picture；
Using matching combination and all features for combining extraction and all coded representations are mismatched to based on deep learning
Coder-decoder neural network model is trained, until loss function value is minimum, training terminates, and obtains trained base
In the coder-decoder neural network model of deep learning.
Coder-decoder neural network model based on deep learning, by training set, study to network parameter, on
Clothing product vector, lower clothing product vector and term vector.
In training process, loss function, comprising: match penalties, generational loss and regularization loss；Wherein,
What match penalties were measured is the order of accuarcy of matching prediction, and it is more accurate to predict, is lost smaller；
What generational loss was measured is that network generates the probability really commented on, and probability is bigger, loses smaller.
Regularization loss is used to the parameter in constraint network, avoids its excessive, the parameter value in network is smaller, and loss is got over
It is small.
Network parameter, jacket article vector, lower clothing product vector and term vector are carried out using back-propagating BP algorithm
It updates to reduce loss.
Therefore, in application embodiment, recommended models are trained using the useful information hidden in user comment, from
And the effect recommended is promoted, while energy analog subscriber generates explanation of the comment as recommendation to the result recommended, improve recommendation
Interpretation.
Second aspect, the disclosure additionally provide the interpretation clothes recommender system of fusion comment；
Merge the interpretation clothes recommender system of comment, comprising:
Model construction module is configured as constructing the coder-decoder neural network model based on deep learning；
Model training module is configured as instructing the coder-decoder neural network model based on deep learning
Practice；
Model uses module, is configured as jacket picture and lower clothing picture to be recommended, while being input to trained
Coder-decoder neural network model, the model give a mark to the matching degree of jacket picture and lower clothing picture, according to beating
Divide sequence to provide recommendation results, while providing the simulation comment of matching degree.
Therefore, in application embodiment, recommended models are trained using the useful information hidden in user comment, from
And the effect recommended is promoted, while energy analog subscriber generates explanation of the comment as recommendation to the result recommended, improve recommendation
Interpretation.
The third aspect, the disclosure additionally provide a kind of electronic equipment, including memory and processor and are stored in storage
The computer instruction run on device and on a processor when the computer instruction is run by processor, is completed first aspect and is appointed
Method in one possible implementation.
Fourth aspect, the disclosure additionally provide a kind of computer readable storage medium, described for storing computer instruction
When computer instruction is executed by processor, in the completion any possible implementation of first aspect the step of method.
Compared with prior art, the beneficial effect of the disclosure is:
The present invention is directed to generate to promote the interpretation of the effect of clothes recommendation and clothes recommendation in conjunction with comment.This hair
It is bright relative to past clothes recommended method, since the useful information being utilized in user comment carrys out training pattern, thus taking
It fills and all achieves good promotion on multiple evaluation metrics in recommendation field.Simultaneously the present invention can recommend while analog subscriber
Comment is generated, the interpretation of recommendation is substantially increased, allows recommender system to become more transparent, credible, can also user be helped to do
Faster out, better decision.
Detailed description of the invention
The accompanying drawings constituting a part of this application is used to provide further understanding of the present application, and the application's shows
Meaning property embodiment and its explanation are not constituted an undue limitation on the present application for explaining the application.
Fig. 1 is the work flow diagram of the NOR of one or more embodiments；
Fig. 2 is the jacket encoder and lower clothing encoder of one or more embodiments；
Fig. 3 is the matching and decoding device of one or more embodiments；
Fig. 4 is the generation decoder of one or more embodiments.
Specific embodiment
It is noted that described further below be all exemplary, it is intended to provide further instruction to the application.Unless another
It indicates, all technical and scientific terms used herein has usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
Embodiment one:
Present invention employs the currently a popular coder-decoder frame based on deep learning, entitled Neural
Outfit Recommendation (referred to as NOR), it comprises three parts: jacket and lower clothing encoder, matching and decoding device
With generation decoder.Wherein jacket and lower clothing encoder are for extracting visual signature from jacket and lower clothing picture.We are upper
Clothing and lower clothing encoder propose a kind of interactive attention mechanism, the match information between jacket and lower clothing can be encoded into institute
In the visual signature of extraction.Matching and decoding device one marking obtained based on extracted visual signature come assess given jacket and
Matching degree between candidate lower clothing.Generating decoder can use extracted visual signature then to generate a word conduct
The comment that the jacket and lower clothing are combined.We propose a kind of cross-module state attention mechanism to generate decoder, can be more
Visual signature is effectively utilized to generate each word.The work flow diagram of NOR is as shown in Figure 1.
The various pieces of NOR are described in detail below.
1. jacket and lower clothing encoder
Jacket encoder uses two structures identical with lower clothing encoder, the convolutional neural networks of parameter sharing
(abbreviation CNN), workflow are as shown in Figure 2；
The jacket of input and lower clothing picture first pass around two layers of convolutional layer, extract visual signature, then we this two
Along channel, this axis is stitched together layer convolution feature, and finally using one layer of pond layer, obtained feature is denoted as WithWherein L is characteristic, and D is the dimension of feature.
We apply a kind of interactive attention mechanism and weave the match information between jacket picture and lower clothing picture later
Code is into the feature extracted.Here for calculating jacket picture to the attention weight of lower clothing picture.We are with entirely first
Office Chi Hualai calculates the global characteristics g of jacket picturet∈RD, as shown in formula (1):
Indicate the ith feature of jacket picture；
Then rightWe calculate g with formula (2)tTo its attention weight et,i:
Wherein WaAnd Ua∈RD×D, va∈RDIt is the parameter in network.Then we need to et,iIt is normalized:
Finally we add the attention weight of lower clothing picture to the visual signature of lower clothing picture with jacket picture
Power summation is to obtain the attention global characteristics of lower clothing picture
Indicate the ith feature of lower clothing picture；
We calculate lower clothing picture to the attention weight of jacket picture in the same way, and obtain jacket picture
Attention global characteristics, are denoted asThen, weWithFurther it is mapped to two visual feature vectorsWithAs shown in formula (5):
WhereinIt is the parameter in network.It is useful in order to learn from the history match of fashion accessory record
Information, we are also that each jacket and lower clothing have learnt an article vector and indicated, are denoted asWithWe will regard
Feel that feature vector and article vector are stitched together the coded representation v final as jacket and lower clothingtAnd vb∈Rm, such as formula (6) institute
Show:
Wherein m=2mv。
2. matching and decoding device
We are the coded representation v based on obtained jacket and lower clothingtAnd vb, with multi-layer perception (MLP) (referred to as MLP) come pre-
The matching marking between given jacket and lower clothing is surveyed, as shown in Figure 3:
Shown in specific mathematical procedure such as formula (7) and formula (8):
hr=ReLU (Wsvt+Usvb) (7)
p(rtb)=softmax (Wrhr) (8)
Wherein hr∈Rn, Ws and Us∈Rn×m, Wr∈R2×nIt is the parameter in network.P (the r finally exportedtb) it is one
Probability distribution, corresponding p (rtb=0) and p (rtb=1), wherein rtb=1 indicates that given jacket and lower clothing match, and rtb=0 table
Show that given jacket and lower clothing mismatch.We are by the matching degree of jacket and lower clothing, that is, p (rtb=1) it is considered as matching marking.
3. generating decoder
For the combination producing comment to given jacket and lower clothing, we used gating cycle neural networks (referred to as
It is GRU) as encoder is generated, as shown in Figure 4:
We calculate the original state s of GRU with the coded representation of jacket and lower clothing first0∈Rq, as shown in formula (9):
s0=tanh (Wivt+Uivb) (9)
Wherein WiAnd Ut∈Rq×mIt is the parameter in network.Each time step τ later, we are to output before GRU input
Word term vector wτ-1∈Re, current context vector ctxτ∈RDState s beforeτ-1∈RqTo calculate new shape
State sτWith current output oτ∈Rq, as shown in formula (10):
sτ,oτ=GRU (wτ-1,ctxτ,sτ-1) (10)
Wherein context vector ctxτ, by it is proposed that cross-module state attention mechanism calculate.Specifically, we will
The jacket visual signature and lower clothing visual signature extracted is combined, and is obtainedThen
We calculate ctx to formula (13) by formula (11)τ:
Wherein Wg∈Rq×DIt is the parameter in network.By cross-module state attention mechanism, we allow generation decoder can be note
Meaning power, which is placed on effective visual signature, guarantees making full use of to the visual signature of extraction.Finally we are come pre- by formula (14)
It surveys current time and walks the word to be generated:
p(wτ|w1,…,wτ-1)=softmax (Wooτ+Uoctxτ) (14)
Wherein Wo∈R|V|×q, Uo∈R|V|×D, V is our dictionary.p(wτ|w1,…,wτ-1) what is returned is the τ word
wτProbability distribution on entire dictionary, we take the word of maximum probability as current prediction result when prediction.If worked as
Preceding prediction result is fullstop, illustrates to have generated a complete sentence, then the word generated all time steps is sequentially
A sentence is connected into return.
NOR before application, needs to learn network parameter, article vector and term vector on training set.Training set by from
The real user that online fashion community crawls thinks that matched jacket and the combination of lower clothing and user comment are constituted.We are logical simultaneously
Negative sampling technique is crossed to obtain it is considered that unmatched jacket and the combination of lower clothing.We define loss function such as formula (15) and arrive later
Shown in formula (18):
L=Lmat+Lgen+Lreg (18)
Wherein P+It is matching combination of sets, P-It is to mismatch combination of sets, CtbIt is the comment set of matching combination (t, b), Θ is
Whole parameters in network and, LmatCorresponding matching loss, LgenCorresponding generational loss, LregCorresponding regularization loss.Due to not
Matched combination, we do not comment on, so we do not consider generational loss therein.And from these true user comments
In, NOR may learn useful clothing matching information.Then we with the common Back Propagation Algorithm of deep learning (referred to as
For BP algorithm) parameter update is carried out to reduce loss to network.
After NOR training, parameter, article vector and term vector are just all fixed, and then can be used for given upper
Clothing and the marking of lower clothing prediction and matching and comment generate.In clothing under recommending for given jacket, we are first with NOR to candidate
Each of lower clothing calculate it and the matching of jacket is given a mark, then sorted lower clothing to obtain recommendation results according to score height.
NOR also generates the reasons why comment is as recommending simultaneously.Recommend jacket similarly such for lower clothing.
The present invention is directed to generate to promote the interpretation of the effect of clothes recommendation and clothes recommendation in conjunction with comment.This hair
It is bright relative to past clothes recommended method, since the useful information being utilized in user comment carrys out training pattern, thus taking
It fills and all achieves good promotion on multiple evaluation metrics in recommendation field.Simultaneously the present invention can recommend while analog subscriber
Comment is generated, the interpretation of recommendation is substantially increased, allows recommender system to become more transparent, credible, can also user be helped to do
Faster out, better decision.
Embodiment two:
The disclosure additionally provides the interpretation clothes recommender system of fusion comment；
Merge the interpretation clothes recommender system of comment, comprising:
Model construction module is configured as constructing the coder-decoder neural network model based on deep learning；
Model training module is configured as instructing the coder-decoder neural network model based on deep learning
Practice；
Model uses module, is configured as jacket picture and lower clothing picture to be recommended, while being input to trained
Coder-decoder neural network model, the model give a mark to the matching degree of jacket picture and lower clothing picture, according to beating
Divide sequence to provide recommendation results, while providing the simulation comment of matching degree.
Embodiment three:
The disclosure additionally provides a kind of electronic equipment, including memory and processor and storage on a memory and are being located
The computer instruction that runs on reason device, when the computer instruction is run by processor, each operation in Method Of Accomplishment, in order to
Succinctly, details are not described herein.
It should be understood that in the disclosure, which can be central processing unit CPU, which, which can be said to be, can be it
His general processor, digital signal processor DSP, application-specific integrated circuit ASIC, ready-made programmable gate array FPGA or other
Programmable logic device, discrete gate or transistor logic, discrete hardware components etc..General processor can be micro process
Device or the processor are also possible to any conventional processor etc..
The memory may include read-only memory and random access memory, and to processor provide instruction and data,
The a part of of memory can also include non-volatile RAM.For example, memory can be with the letter of storage device type
Breath.
During realization, each step of the above method can by the integrated logic circuit of the hardware in processor or
The instruction of software form is completed.The step of method in conjunction with disclosed in the disclosure, can be embodied directly in hardware processor and execute
At, or in processor hardware and software module combination execute completion.Software module can be located at random access memory, dodge
It deposits, this fields are mature deposits for read-only memory, programmable read only memory or electrically erasable programmable memory, register etc.
In storage media.The storage medium is located at memory, and processor reads the information in memory, completes the above method in conjunction with its hardware
The step of.To avoid repeating, it is not detailed herein.Those of ordinary skill in the art may be aware that in conjunction with institute herein
Each exemplary unit, that is, algorithm steps of disclosed embodiment description, can be hard with electronic hardware or computer software and electronics
The combination of part is realized.These functions are implemented in hardware or software actually, the specific application depending on technical solution
And design constraint.Professional technician can realize described function using distinct methods to each specific application
Can, but this realization is it is not considered that exceed scope of the present application.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In several embodiments provided herein, it should be understood that disclosed systems, devices and methods, it can be with
It realizes in other way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit
It divides, only a kind of division of logic function, there may be another division manner in actual implementation, such as multiple units or group
Part can be combined or can be integrated into another system, or some features can be ignored or not executed.In addition, showing
The mutual coupling or direct-coupling or communication connection shown or discussed can be through some interfaces, device or unit
Indirect coupling or communication connection, can be electrically, mechanical or other forms.
Example IV:
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer readable storage medium.Based on this understanding, the technical solution of the application is substantially right in other words
The part of part or the technical solution that the prior art contributes can be embodied in the form of software products, the calculating
Machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be individual
Computer, server or network equipment etc.) execute each embodiment the method for the application all or part of the steps.And it is preceding
The storage medium stated includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory
The various media that can store program code such as (RAM, Random Access Memory), magnetic or disk.
The foregoing is merely preferred embodiment of the present application, are not intended to limit this application, for the skill of this field
For art personnel, various changes and changes are possible in this application.Within the spirit and principles of this application, made any to repair
Change, equivalent replacement, improvement etc., should be included within the scope of protection of this application.
Claims (10)
1. merging the interpretation clothes recommended method of comment, characterized in that include:
Construct the coder-decoder neural network model based on deep learning；
Coder-decoder neural network model based on deep learning is trained；
By jacket picture and lower clothing picture to be recommended, while it being input to trained coder-decoder neural network model,
The model gives a mark to the matching degree of jacket picture and lower clothing picture, provides recommendation results according to marking sequence, gives simultaneously
The simulation comment of matching degree out.
2. the method as described in claim 1, characterized in that the coder-decoder neural network based on deep learning
Model, comprising:
Jacket encoder, lower clothing encoder, matching and decoding device and generation decoder；
The jacket encoder extracts the jacket visual signature and jacket coding schedule of jacket picture for receiving jacket picture
Show；The jacket coded representation includes the match information between jacket picture and lower clothing picture；
The lower clothing encoder extracts the lower clothing visual signature and lower clothing coding schedule of lower clothing picture for receiving lower clothing picture
Show；The lower clothing coded representation includes the match information between jacket picture and lower clothing picture；
The matching and decoding device is used for according to jacket coded representation and lower clothing coded representation, between jacket picture and lower clothing picture
Matching degree give a mark；
The generation decoder is used for according to jacket visual signature, jacket coded representation, lower clothing visual signature and lower clothing coding schedule
Show, comment is simulated to the combination producing of jacket picture and lower clothing picture.
3. method according to claim 2, characterized in that extract the specific steps of the jacket visual signature of jacket picture are as follows:
The jacket encoder, comprising: sequentially connected first convolutional layer, the second convolutional layer, the first splicing layer and the first pond
Change layer；
First convolutional layer carries out Visual Feature Retrieval Process to jacket picture, obtains First look feature；
Second convolutional layer carries out Visual Feature Retrieval Process to jacket picture, obtains the second visual signature；
The first splicing layer carries out series connection splicing to First look feature and the second visual signature, and the third that splicing is obtained regards
Feel that feature is sent into the first pond layer；
First pond layer handles third visual signature, obtains the jacket visual signature of jacket picture；
Alternatively,
Extract the specific steps of the lower clothing visual signature of lower clothing picture are as follows:
The lower clothing encoder, comprising: sequentially connected third convolutional layer, Volume Four lamination, the second splicing layer and the second pond
Change layer；
The third convolutional layer carries out Visual Feature Retrieval Process to lower clothing picture, obtains the 4th visual signature；
The Volume Four lamination carries out Visual Feature Retrieval Process to lower clothing picture, obtains the 5th visual signature；
The second splicing layer carries out series connection splicing to the 4th visual signature and the 5th visual signature, the 6th will obtained after splicing
Visual signature is sent into the second pond layer；
Second pond layer handles the 6th visual signature, obtains the visual signature of lower clothing picture.
4. method according to claim 2, characterized in that extract the specific steps of the jacket coded representation of jacket picture are as follows:
Using interaction attention mechanism, the match information between jacket picture and lower clothing picture is encoded to the jacket picture of extraction
Visual signature in, obtain the coded representation of jacket picture；
Alternatively,
Extract the specific steps of the lower clothing coded representation of lower clothing picture are as follows:
Using interaction attention mechanism, the match information between jacket picture and lower clothing picture is encoded to the lower clothing picture of extraction
Visual signature in, obtain the coded representation of lower clothing picture.
5. method as claimed in claim 4, characterized in that using interaction attention mechanism, by jacket picture and lower clothing picture
Between match information be encoded in the visual signature of jacket picture of extraction, obtain the specific step of the coded representation of jacket picture
Suddenly are as follows:
Firstly, obtaining the global characteristics of lower clothing picture by the average value for calculating lower clothing visual signature；
Then, to each visual signature of jacket picture, visual signature of the global characteristics to jacket picture of lower clothing picture is calculated
Attention weight；Attention weight is normalized；
Secondly, using the global characteristics of lower clothing picture to the attention weight of the visual signature of jacket picture, to jacket picture
Visual signature is weighted summation, obtains the attention global characteristics of jacket picture；
Again, the attention global characteristics of jacket picture are mapped to visual feature vector；
Again, the visual feature vector of jacket picture jacket article vector corresponding with jacket picture is subjected to series connection splicing, spelled
The result being connected to is the coded representation of final jacket picture；
Alternatively,
Using interaction attention mechanism, the match information between jacket picture and lower clothing picture is encoded to the lower clothing picture of extraction
Visual signature in, obtain the specific steps of the coded representation of lower clothing picture are as follows:
Firstly, obtaining the global characteristics of jacket picture by the average value for calculating jacket visual signature；
Then, to each visual signature of lower clothing picture, the global characteristics for calculating jacket picture are special to the vision of lower clothing picture
The attention weight of sign；Attention weight is normalized；
Secondly, using the global characteristics of jacket picture to the attention weight of the visual signature of lower clothing picture, to lower clothing picture
Visual signature is weighted summation, obtains the attention global characteristics of lower clothing picture；
Again, the attention global characteristics of lower clothing picture are mapped to visual feature vector；
Again, the visual feature vector of lower clothing picture lower clothing product vector corresponding with lower clothing picture is subjected to series connection splicing, spelled
The result being connected to is the coded representation of final lower clothing picture.
6. method as claimed in claim 5, characterized in that the obtaining step of jacket article vector are as follows:
Firstly, one jacket article vector matrix of random initializtion, the corresponding jacket of each of these row；
Then, according to the jacket picture of input, corresponding vector is obtained from jacket article vector matrix, based on later
It calculates；
Finally, jacket article vector matrix is by together with the parameter of neural network, with the minimum target of loss function value, by rear
It is updated to BP algorithm is propagated, finally obtains updated jacket article vector；
Alternatively,
The obtaining step of lower clothing product vector are as follows:
Firstly, random initializtion one lower clothing product vector matrix, the corresponding lower clothing of each of these row；
Then, according to the lower clothing picture of input, corresponding vector is obtained from lower clothing product vector matrix, based on later
It calculates；
Finally, lower clothing product vector matrix is by together with the parameter of neural network, with the minimum target of loss function value, by rear
It is updated to BP algorithm is propagated, finally obtains updated lower clothing product vector.
7. method according to claim 2, characterized in that according to jacket coded representation and lower clothing coded representation, to jacket figure
The specific steps that matching degree between piece and lower clothing picture is given a mark are as follows:
It using jacket coded representation and lower clothing coded representation as input value, is input in MLP multi-layer perception (MLP), output is jacket
The matching marking result of picture and lower clothing picture；
Alternatively,
According to jacket visual signature, jacket coded representation, lower clothing visual signature and lower clothing coded representation, to jacket picture and lower clothing
The combination producing of picture simulates the step of comment are as follows:
Step (1): building gating cycle neural network GRU；
Step (2): using the coded representation of jacket and lower clothing, the original state of gating cycle neural network GRU is calculated；
Step (3): the circulate operation that gating cycle neural network GRU carries out step (31) to step (33) is complete until generating one
Whole sentence:
Step (31): it first to jacket visual signature and lower clothing visual signature, is handled to obtain using cross-module state attention mechanism
The context vector of current time step；
Step (32): by the state of a upper time step of gating cycle neural network GRU, a upper time step generates the word of word
The context vector of vector sum current time step is input in gating cycle neural network GRU, obtains the new shape of current time step
State and to be currently generated word prediction probability be distributed；
Step (33): the word of maximum probability is chosen as current generation result；The word includes punctuation mark；If worked as
Preceding generation result is fullstop, illustrates to have generated a complete sentence, then the word generated all time steps is sequentially
A sentence is connected into return；
Alternatively,
To jacket visual signature and lower clothing visual signature, handled to obtain current time step using cross-module state attention mechanism
The specific steps of context vector are as follows:
Firstly, corresponding tandem compound to jacket visual signature and lower clothing visual signature；
Then, the state of the upper time step of gating cycle neural network GRU is calculated to jacket visual signature and lower clothing vision
After feature tandem compound, each combined attention weight；
Then, using the attention weight of calculating, weighted sum is done to all tandem compounds, the result finally returned to be exactly when
The context vector of preceding time step；
Alternatively,
The acquisition modes of term vector are:
Firstly, one term vector matrix of random initializtion, the corresponding word of each of these row；
Then, according to word currently entered, corresponding vector is obtained from term vector matrix, for calculating later；
Finally, term vector matrix, with the minimum target of loss function, passes through back-propagating BP for together with the parameter of neural network
Algorithm is updated；
Alternatively,
The specific steps that coder-decoder neural network model based on deep learning is trained are as follows:
Training set includes the combination of the matched jacket that the real user crawled from online fashion community website provides and lower clothing,
Each combination includes jacket picture, lower clothing picture, thumbs up several and user comment；
Combination of the number greater than threshold value will be thumbed up and be considered as matching combination；Then combination is mismatched to obtain by negative sampling, i.e., at random
It chooses a jacket and a lower clothing constitutes a combination, if the combination does not occur in matching combination, which is regarded
To mismatch combination；To the jacket picture and lower clothing picture in matching combination, visual signature, the jacket of jacket picture are extracted respectively
The coded representation of picture, the coded representation of the visual signature of lower clothing picture and lower clothing picture；
To the jacket picture and lower clothing picture mismatched in combination, the visual signature of jacket picture, jacket picture are respectively extracted
Coded representation, the coded representation of the visual signature of lower clothing picture and lower clothing picture；
Using matching combination and all features for combining extraction and all coded representations are mismatched to the coding based on deep learning
Device-decoder neural network model is trained, until loss function value is minimum, training terminates, and is obtained trained based on deep
Spend the coder-decoder neural network model of study.
8. merging the interpretation clothes recommender system of comment, characterized in that include:
Model construction module is configured as constructing the coder-decoder neural network model based on deep learning；
Model training module is configured as being trained the coder-decoder neural network model based on deep learning；
Model uses module, is configured as jacket picture and lower clothing picture to be recommended, while being input to trained coding
Device-decoder neural network model, the model give a mark to the matching degree of jacket picture and lower clothing picture, are arranged according to marking
Sequence provides recommendation results, while providing the simulation comment of matching degree.
9. a kind of electronic equipment, characterized in that on a memory and on a processor including memory and processor and storage
The computer instruction of operation when the computer instruction is run by processor, is completed described in any one of claim 1-7 method
Step.
10. a kind of computer readable storage medium, characterized in that for storing computer instruction, the computer instruction is located
When managing device execution, step described in any one of claim 1-7 method is completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910024347.2A CN109754317B (en) | 2019-01-10 | 2019-01-10 | Comment-fused interpretable garment recommendation method, system, device and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910024347.2A CN109754317B (en) | 2019-01-10 | 2019-01-10 | Comment-fused interpretable garment recommendation method, system, device and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109754317A true CN109754317A (en) | 2019-05-14 |
| CN109754317B CN109754317B (en) | 2020-11-06 |
Family
ID=66405439
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910024347.2A Active CN109754317B (en) | 2019-01-10 | 2019-01-10 | Comment-fused interpretable garment recommendation method, system, device and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109754317B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110188449A (en) * | 2019-05-27 | 2019-08-30 | 山东大学 | Interpretable clothing information recommended method, system, medium and equipment based on attribute |
| CN110321473A (en) * | 2019-05-21 | 2019-10-11 | 山东省计算中心（国家超级计算济南中心） | Diversity preference information method for pushing, system, medium and equipment based on multi-modal attention |
| CN110688832A (en) * | 2019-10-10 | 2020-01-14 | 河北省讯飞人工智能研究院 | Comment generation method, device, equipment and storage medium |
| CN110765353A (en) * | 2019-10-16 | 2020-02-07 | 腾讯科技（深圳）有限公司 | Processing method and device of project recommendation model, computer equipment and storage medium |
| CN110807477A (en) * | 2019-10-18 | 2020-02-18 | 山东大学 | Attention mechanism-based neural network garment matching scheme generation method and system |
| CN111046286A (en) * | 2019-12-12 | 2020-04-21 | 腾讯科技（深圳）有限公司 | Object recommendation method and device and computer storage medium |
| CN111400525A (en) * | 2020-03-20 | 2020-07-10 | 中国科学技术大学 | Intelligent fashionable garment matching and recommending method based on visual combination relation learning |
| CN111476622A (en) * | 2019-11-21 | 2020-07-31 | 北京沃东天骏信息技术有限公司 | Article pushing method and device and computer readable storage medium |
| CN113158045A (en) * | 2021-04-20 | 2021-07-23 | 中国科学院深圳先进技术研究院 | Interpretable recommendation method based on graph neural network reasoning |
| CN113850656A (en) * | 2021-11-15 | 2021-12-28 | 内蒙古工业大学 | Personalized clothing recommendation method and system based on attention perception and integrating multi-mode data |
| CN117994007A (en) * | 2024-04-03 | 2024-05-07 | 山东科技大学 | Social recommendation method based on multi-view fusion heterogeneous graph neural network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140129371A1 (en) * | 2012-11-05 | 2014-05-08 | Nathan R. Wilson | Systems and methods for providing enhanced neural network genesis and recommendations |
| US20150339757A1 (en) * | 2014-05-20 | 2015-11-26 | Parham Aarabi | Method, system and computer program product for generating recommendations for products and treatments |
| CN106815739A (en) * | 2015-12-01 | 2017-06-09 | 东莞酷派软件技术有限公司 | A kind of recommendation method of clothing, device and mobile terminal |
| CN107590584A (en) * | 2017-08-14 | 2018-01-16 | 上海爱优威软件开发有限公司 | Dressing collocation reviewing method |
| CN107993131A (en) * | 2017-12-27 | 2018-05-04 | 广东欧珀移动通信有限公司 | Wear to take and recommend method, apparatus, server and storage medium |
| CN108734557A (en) * | 2018-05-18 | 2018-11-02 | 北京京东尚科信息技术有限公司 | Methods, devices and systems for generating dress ornament recommendation information |
| CN109117779A (en) * | 2018-08-06 | 2019-01-01 | 百度在线网络技术（北京）有限公司 | One kind, which is worn, takes recommended method, device and electronic equipment |
-
2019
- 2019-01-10 CN CN201910024347.2A patent/CN109754317B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140129371A1 (en) * | 2012-11-05 | 2014-05-08 | Nathan R. Wilson | Systems and methods for providing enhanced neural network genesis and recommendations |
| US20150339757A1 (en) * | 2014-05-20 | 2015-11-26 | Parham Aarabi | Method, system and computer program product for generating recommendations for products and treatments |
| CN106815739A (en) * | 2015-12-01 | 2017-06-09 | 东莞酷派软件技术有限公司 | A kind of recommendation method of clothing, device and mobile terminal |
| CN107590584A (en) * | 2017-08-14 | 2018-01-16 | 上海爱优威软件开发有限公司 | Dressing collocation reviewing method |
| CN107993131A (en) * | 2017-12-27 | 2018-05-04 | 广东欧珀移动通信有限公司 | Wear to take and recommend method, apparatus, server and storage medium |
| CN108734557A (en) * | 2018-05-18 | 2018-11-02 | 北京京东尚科信息技术有限公司 | Methods, devices and systems for generating dress ornament recommendation information |
| CN109117779A (en) * | 2018-08-06 | 2019-01-01 | 百度在线网络技术（北京）有限公司 | One kind, which is worn, takes recommended method, device and electronic equipment |
Non-Patent Citations (2)
| Title |
|---|
| JIANMO NI: "Estimating Reactions and Recommending Products with Generative Models of Reviews", 《PROCEEDINGS OF THE THE 8TH INTERNATIONAL JOINT CONFERENCE ON NATURAL LANGUAGE PROCESSING》 * |
| 金泰伟: "基于用户行为数据和评论数据的推荐模型研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110321473B (en) * | 2019-05-21 | 2021-05-25 | 山东省计算中心（国家超级计算济南中心） | Multi-modal attention-based diversity preference information pushing method, system, medium and device |
| CN110321473A (en) * | 2019-05-21 | 2019-10-11 | 山东省计算中心（国家超级计算济南中心） | Diversity preference information method for pushing, system, medium and equipment based on multi-modal attention |
| CN110188449A (en) * | 2019-05-27 | 2019-08-30 | 山东大学 | Interpretable clothing information recommended method, system, medium and equipment based on attribute |
| CN110688832A (en) * | 2019-10-10 | 2020-01-14 | 河北省讯飞人工智能研究院 | Comment generation method, device, equipment and storage medium |
| CN110688832B (en) * | 2019-10-10 | 2023-06-09 | 河北省讯飞人工智能研究院 | Comment generation method, comment generation device, comment generation equipment and storage medium |
| CN110765353A (en) * | 2019-10-16 | 2020-02-07 | 腾讯科技（深圳）有限公司 | Processing method and device of project recommendation model, computer equipment and storage medium |
| CN110765353B (en) * | 2019-10-16 | 2022-03-08 | 腾讯科技（深圳）有限公司 | Processing method and device of project recommendation model, computer equipment and storage medium |
| CN110807477A (en) * | 2019-10-18 | 2020-02-18 | 山东大学 | Attention mechanism-based neural network garment matching scheme generation method and system |
| CN110807477B (en) * | 2019-10-18 | 2022-06-07 | 山东大学 | Attention mechanism-based neural network garment matching scheme generation method and system |
| CN111476622B (en) * | 2019-11-21 | 2021-05-25 | 北京沃东天骏信息技术有限公司 | Article pushing method and device and computer readable storage medium |
| CN111476622A (en) * | 2019-11-21 | 2020-07-31 | 北京沃东天骏信息技术有限公司 | Article pushing method and device and computer readable storage medium |
| CN111046286B (en) * | 2019-12-12 | 2023-04-18 | 腾讯科技（深圳）有限公司 | Object recommendation method and device and computer storage medium |
| CN111046286A (en) * | 2019-12-12 | 2020-04-21 | 腾讯科技（深圳）有限公司 | Object recommendation method and device and computer storage medium |
| CN111400525A (en) * | 2020-03-20 | 2020-07-10 | 中国科学技术大学 | Intelligent fashionable garment matching and recommending method based on visual combination relation learning |
| CN111400525B (en) * | 2020-03-20 | 2023-06-16 | 中国科学技术大学 | Fashion clothing intelligent matching and recommending method based on vision combination relation learning |
| CN113158045A (en) * | 2021-04-20 | 2021-07-23 | 中国科学院深圳先进技术研究院 | Interpretable recommendation method based on graph neural network reasoning |
| CN113158045B (en) * | 2021-04-20 | 2022-11-01 | 中国科学院深圳先进技术研究院 | Interpretable recommendation method based on graph neural network reasoning |
| CN113850656A (en) * | 2021-11-15 | 2021-12-28 | 内蒙古工业大学 | Personalized clothing recommendation method and system based on attention perception and integrating multi-mode data |
| CN117994007A (en) * | 2024-04-03 | 2024-05-07 | 山东科技大学 | Social recommendation method based on multi-view fusion heterogeneous graph neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109754317B (en) | 2020-11-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109754317A (en) | Merge interpretation clothes recommended method, system, equipment and the medium of comment | |
| CN109657156A (en) | A kind of personalized recommendation method generating confrontation network based on circulation | |
| Song et al. | A novel visible-depth-thermal image dataset of salient object detection for robotic visual perception | |
| CN109684478A (en) | Disaggregated model training method, classification method and device, equipment and medium | |
| CN110060097A (en) | User behavior sequence of recommendation method based on attention mechanism and convolutional neural networks | |
| CN110008408A (en) | A kind of session recommended method, system, equipment and medium | |
| CN108960959A (en) | Multi-modal complementary garment coordination method, system and medium neural network based | |
| CN107886089A (en) | A kind of method of the 3 D human body Attitude estimation returned based on skeleton drawing | |
| CN108875910A (en) | Garment coordination method, system and the storage medium extracted based on attention knowledge | |
| CN109783539A (en) | Usage mining and its model building method, device and computer equipment | |
| Wu et al. | ClothGAN: generation of fashionable Dunhuang clothes using generative adversarial networks | |
| CN110955826A (en) | Recommendation system based on improved recurrent neural network unit | |
| CN108984555A (en) | User Status is excavated and information recommendation method, device and equipment | |
| CN109871736A (en) | The generation method and device of natural language description information | |
| CN110223358A (en) | Visible pattern design method, training method, device, system and storage medium | |
| CN110008999A (en) | Determination method, apparatus, storage medium and the electronic device of target account number | |
| Hsiao et al. | A study on the application of an artificial neural algorithm in the color matching of Taiwanese cultural and creative commodities | |
| CN109117943A (en) | Utilize the method for more attribute informations enhancing network characterisation study | |
| CN109978074A (en) | Image aesthetic feeling and emotion joint classification method and system based on depth multi-task learning | |
| Cao et al. | Explainable high-order visual question reasoning: A new benchmark and knowledge-routed network | |
| CN110209860A (en) | A kind of interpretable garment coordination method and device based on clothes attribute of template-directed | |
| CN112330362A (en) | Rapid data intelligent analysis method for internet mall user behavior habits | |
| Sun et al. | Locate: End-to-end localization of actions in 3d with transformers | |
| CN111728302A (en) | Garment design method and device | |
| CN110210523A (en) | A kind of model based on shape constraint diagram wears clothing image generating method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |